The family is separable in trace norm
ProvedBosonicLearn.fugacityFamily_separableThe hard family has a countable subfamily that is dense in trace norm.
Let and . There is a sequence such that for every and every there is with
Why it holds. is a compact subset of the finite-dimensional space , hence separable, and is continuous in trace norm (photon-number cutoff).
Role. In the proof of Theorem 5 an estimate of the state is converted into an estimate of by taking a nearest point of a net. With a countable dense family the nearest-point rule "first index with " is a measurable map from states to , which is what Proposition 5 requires of the estimator.
Formalization Note is traceSize on the coefficient arrays fugacityCoeff (fugacityFamily ν ξ X). The paper uses a finite -net for ; the countable family valid for all is used instead so that no compactness argument is needed downstream.
import Mathlib import Definitions.Def_BosonicLearn_fugacityCoeff import Definitions.Def_BosonicLearn_fugacityFamily import Definitions.Def_BosonicLearn_Gsupp open QAlgorithms.Papers.Independent.Dependency50.P33 (State traceSize) open scoped ENNReal
namespace BosonicLearn
/-- **The hard family is separable in trace norm.**
Chen, Anna Mele, Anna Mele, Preskill, *When are bosonic Gaussian states classical to learn?*, arXiv:2609.26705, proof of Theorem 5, Eq. (106), p. 40: "Since G_supp is compact and the map X ↦ ρ_{R_X} is continuous in
trace norm (which can be shown by appropriate photon number cutoff), the image {ρ_{R_X} : X ∈ G_supp} has an
η-covering net in trace norm".
For `0 < ν ≤ 1` and `0 < ξ < 1/32` there is a sequence `D₀, D₁, … ∈ G_supp` such that for every `X ∈ G_supp` and
every `η > 0` some `k` has `‖ρ_{R_X} − ρ_{R_{D_k}}‖₁ < η`.
**Formalization Note.** The paper uses a finite `η`-net for one `η`; a countable family that works for every `η`
exists for the same reason (continuity of `X ↦ ρ_{R_X}` in trace norm and separability of `G_supp`) and is what a
measurable nearest-point decoder needs. `‖·‖₁` is `traceSize`. -/
theorem fugacityFamily_separable (n : ℕ) (ν ξ : ℝ) (hν0 : 0 < ν) (hν1 : ν ≤ 1) (hξ0 : 0 < ξ)
(hξ1 : ξ < 1 / 32) :
∃ D : ℕ → Herm0 n, (∀ k, D k ∈ Gsupp n) ∧ ∀ X ∈ Gsupp n, ∀ η : ℝ, 0 < η →
∃ k, traceSize (fugacityCoeff (fugacityFamily ν ξ X) -
fugacityCoeff (fugacityFamily ν ξ (D k))) < ENNReal.ofReal η := by sorry
end BosonicLearn